Beyond the AI layer: Why becoming an AI-first bank means redesigning banking itself
Banks have spent years adding AI to their digital estates: chatbots in the contact centre, copilots for relationship managers, models in the credit workflow. Almost every institution now has AI projects. Far fewer have an AI operating model
In partnership with
VeriPark
VeriPark is a global solutions provider enabling financial institutions to become digital leaders by placing Customer Experience at the core of digital transformation. With offices located in the United Kingdom, Europe, Canada, Asia, Africa and the Middle East, VeriPark is helping financial institutions to enhance their customer acquisition, retention and...
Banks have spent years adding AI to their digital estates: chatbots in the contact centre, copilots for relationship managers, models in the credit workflow. Almost every institution now has AI projects. Far fewer have an AI operating model. A recent McKinsey cross-industry survey, cited during the session, shows the gap: 44% of respondents say AI is scaling across their enterprise, but only 37% attribute any contribution to EBIT to it. For financial institutions, the central question is shifting from "Where can we add AI?" to "Which customer journeys, decisions and processes should be redesigned around AI from the start?"
This shift framed a recent Qorus AI Community webinar delivered in partnership with VeriPark. Moderated by Jan Uriga (Senior Advisor at Qorus), it opened with a keynote from Gagan Sethi, Vice President, Product at VeriPark, followed by a panel of three leaders:
• Javier Martínez Rodriguez, Director of Analytics & AI at Banco Sabadell (one of Spain's largest banking groups)
• Emel Çuhacı, Chief Business Development Officer at NEOHUB (DenizBank Group's innovation hub, where banking, startups and technology meet)
• Funda Çınar, Vice President, Sales at VeriPark (a provider of digital banking and CRM solutions for financial institutions)
1. From assistance to action: The rise of agentic banking
For years, AI in banking mostly meant chatbots and customer service. As Gagan Sethi explained, AI now works in three modes. Banks can adopt them together or one after another. In lending, the progression looks like this:
- Assistive: AI collects and summarises the applicant's demographic and financial data and turns it into usable insight.
- Anticipatory: AI identifies patterns, flags exceptions and recommends the next review in the workflow.
- Agentic: AI coordinates the whole credit workflow, from risk assessment and documentation through approval and disbursement.
The next stage changes more than internal processes. Javier Martínez Rodriguez argued that agentic AI could be the biggest change in customer interaction since mobile banking. Today, customers have to learn the bank's menus and complete processes step by step. Intelligent assistants reverse that: they understand the customer's intent and act on it. Customers are also turning to general-purpose AI assistants to research financial products. In future, the party dealing with the bank may sometimes be an AI agent acting on the customer's behalf. He described this as a possible evolution rather than an established reality, and one that raises open questions of identity, consent, security and trust.
Emel Çuhacı took the argument to its conclusion: in ten years, will customers still need a banking app? If an agent can act for them, why should they navigate five separate journeys? For Çuhacı, this is a business-model question rather than a technology one. It requires journeys rebuilt from scratch and new skills and mindsets across the bank.
Sethi set the boundary for all of this. Banking rests on three things: information, decision and trust.
2. Problem first, model second: Where AI is already paying off
The banks seeing measurable returns did not start with a technology mandate. They started with a specific problem. Sethi presented three cases from VeriPark's client base:
- Employee productivity: Tru Cooperative Bank in Canada gave its advisors Microsoft 365 Copilot to gather customer context before conversations. With 93% of employees confident using it and weekly use above 90%, the bank moved on to agents for research, analysis and commercial-banking workflows.
- Customer service: DSK Bank in Bulgaria separated standard requests from those that need human empathy and automated the standard ones with a voice-and-text chatbot. Average answer time fell by more than 80% and handling costs by 19%.
- Risk decisioning: Kuwait Finance House centralised scattered data on its corporate clients and built RiskGPT on top of it. The time to produce a corporate risk rating fell from three days to under an hour, a 96% reduction. The tool was later extended to forecast cash flows and financials.
In each case, the first success led to the next use case. The most important gain was often time returned to people. At DSK, advisors could focus on the customers who needed them. As Jan Uriga observed, the more banking becomes automated, the more valuable real human interaction may become at critical moments. AI should remove low-value work, not the people who handle judgement, empathy and complex decisions.
3. The personalisation paradox: Rich data, thin insight
Customers measure their bank against Amazon, Spotify and Google, and they expect the same relevance. Yet, as Çuhacı pointed out, regulation limits banks to their customers' activity inside the bank. That is still a rich asset: purchases, spending patterns and navigation behaviour can all predict needs. Despite this, Çuhacı said she had yet to see a bank truly invest in understanding that data. Most still rely on propensity models much like those of 20 years ago.
The channels and triggers to reach customers have been in place for years. The gap is insight.
Funda Çınar made the commercial case. Personalisation that covers the bank's whole approach to the customer, not just its product offers, will bring significant revenue growth. She pointed to banks going where their customers spend their time: into retailers' e-commerce journeys, offering personalised loans at the point of purchase.
Trust is the condition for this. Martínez Rodriguez said Banco Sabadell does not expose customer data to public AI tools without a secure process, and uses only tools certified by its security teams.
4. From pilots to production: Scaling AI is an operating-model challenge
Most banks have pilots. Few have turned them into enterprise capabilities. Çuhacı identified three causes:
- IT mindset: Pilots are treated as technology projects rather than business transformations, and tested in sandboxes instead of real lending and compliance workflows.
- Fragmented data: Customer data is scattered across dozens of legacy systems.
- Late governance: Human-in-the-loop validation and risk guardrails are added after the fact rather than designed in from the start.
Sethi's four foundations address each of these: trusted data, orchestration across the wider workflow, governance and human control in proportion to the stakes, and outcome-based ownership. A standard customer query may need no human. A large corporate risk decision does.
Çınar described the economics behind the urgency. Inflation is eroding IT budgets while AI funding grows at double-digit rates, so efficiency is becoming the funding engine for AI. Most failures she has seen trace back to one cause: the benefit was never defined at the start.
For Martínez Rodriguez, scaling depends on three things:
- every initiative tied to a measurable business outcome
- reusable capabilities instead of one-off solutions
- an operating model built on clear ownership, multidisciplinary teams and governance that does not block delivery
5. Strategic outlook
AI is moving from experimentation to the core of how banks operate. As Çuhacı warned, projects that exist just "to make AI things" will not survive that shift. Over the coming years, winning AI strategies will rely on:
- Problem-led investment, with every initiative tied to a defined KPI from day one.
- Journey redesign, rebuilding decisions and processes end to end rather than layering AI on top of them.
- Shared foundations of trusted data, orchestration and built-in governance that make each new use case faster than the last.
Agent-ready trust frameworks for identity and consent, for a future where customers may send an AI agent instead of logging in.
The panel did not settle whether customers will still open a banking app in ten years. On one point, though, they agreed: the next decade of banking will be less about adding intelligence to the bank and more about rebuilding the bank around it. The banks that pull ahead will not be the ones with the longest list of AI pilots. They will be the ones that redesign how decisions are made, how customers are served and how their people create value, while protecting the trust that banking depends on.
The question is no longer whether banks will use AI. They already do. The question is whether they are prepared to redesign the business around it.
AI Community community
With Qorus memberships, you gain access to exclusive innovation best practices and tailored matchmaking opportunities with executives who share your challenges.